3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard.
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 * note, many functions in here may use MMX which trashes the FPU state, it is
27 * absolutely necessary to call emms_c() between dsp & float/double code
30 #ifndef AVCODEC_DSPUTIL_H
31 #define AVCODEC_DSPUTIL_H
38 typedef short DCTELEM;
40 typedef short IDWTELEM;
42 void fdct_ifast (DCTELEM *data);
43 void fdct_ifast248 (DCTELEM *data);
44 void ff_jpeg_fdct_islow (DCTELEM *data);
45 void ff_fdct248_islow (DCTELEM *data);
47 void j_rev_dct (DCTELEM *data);
48 void j_rev_dct4 (DCTELEM *data);
49 void j_rev_dct2 (DCTELEM *data);
50 void j_rev_dct1 (DCTELEM *data);
51 void ff_wmv2_idct_c(DCTELEM *data);
53 void ff_fdct_mmx(DCTELEM *block);
54 void ff_fdct_mmx2(DCTELEM *block);
55 void ff_fdct_sse2(DCTELEM *block);
57 void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
58 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
59 void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
60 void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
61 void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
62 void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
63 void ff_h264_idct_add16_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
64 void ff_h264_idct_add16intra_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
65 void ff_h264_idct8_add4_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
66 void ff_h264_idct_add8_c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
68 void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1,
69 const float *src2, int src3, int blocksize, int step);
70 void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1,
71 const float *win, float add_bias, int len);
72 void ff_float_to_int16_c(int16_t *dst, const float *src, long len);
73 void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels);
76 extern const uint8_t ff_alternate_horizontal_scan[64];
77 extern const uint8_t ff_alternate_vertical_scan[64];
78 extern const uint8_t ff_zigzag_direct[64];
79 extern const uint8_t ff_zigzag248_direct[64];
81 /* pixel operations */
82 #define MAX_NEG_CROP 1024
85 extern uint32_t ff_squareTbl[512];
86 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
88 /* VP3 DSP functions */
89 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
90 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
91 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
93 void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
94 void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values);
96 /* 1/2^n downscaling functions from imgconvert.c */
97 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
98 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
99 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
100 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
102 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
103 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
105 /* minimum alignment rules ;)
106 If you notice errors in the align stuff, need more alignment for some ASM code
107 for some CPU or need to use a function with less aligned data then send a mail
108 to the ffmpeg-devel mailing list, ...
110 !warning These alignments might not match reality, (missing attribute((align))
111 stuff somewhere possible).
112 I (Michael) did not check them, these are just the alignments which I think
113 could be reached easily ...
115 !future video codecs might need functions with less strict alignment
119 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
120 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
121 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
122 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
123 void clear_blocks_c(DCTELEM *blocks);
126 /* add and put pixel (decoding) */
127 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
128 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
129 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
130 typedef void (*tpel_mc_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int w, int h);
131 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
132 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
133 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
134 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
136 #define DEF_OLD_QPEL(name)\
137 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
138 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
139 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
141 DEF_OLD_QPEL(qpel16_mc11_old_c)
142 DEF_OLD_QPEL(qpel16_mc31_old_c)
143 DEF_OLD_QPEL(qpel16_mc12_old_c)
144 DEF_OLD_QPEL(qpel16_mc32_old_c)
145 DEF_OLD_QPEL(qpel16_mc13_old_c)
146 DEF_OLD_QPEL(qpel16_mc33_old_c)
147 DEF_OLD_QPEL(qpel8_mc11_old_c)
148 DEF_OLD_QPEL(qpel8_mc31_old_c)
149 DEF_OLD_QPEL(qpel8_mc12_old_c)
150 DEF_OLD_QPEL(qpel8_mc32_old_c)
151 DEF_OLD_QPEL(qpel8_mc13_old_c)
152 DEF_OLD_QPEL(qpel8_mc33_old_c)
154 #define CALL_2X_PIXELS(a, b, n)\
155 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
156 b(block , pixels , line_size, h);\
157 b(block+n, pixels+n, line_size, h);\
160 /* motion estimation */
161 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
162 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
163 typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size, int h)/* __attribute__ ((const))*/;
167 typedef struct slice_buffer_s slice_buffer;
172 typedef struct ScanTable{
173 const uint8_t *scantable;
174 uint8_t permutated[64];
175 uint8_t raster_end[64];
177 /** Used by dct_quantize_altivec to find last-non-zero */
178 DECLARE_ALIGNED(16, uint8_t, inverse[64]);
182 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
184 void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize,
185 int block_w, int block_h,
186 int src_x, int src_y, int w, int h);
191 typedef struct DSPContext {
192 /* pixel ops : interface with DCT */
193 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
194 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
195 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
196 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
197 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
198 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
199 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
200 int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
202 * translational global motion compensation.
204 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
206 * global motion compensation.
208 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
209 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
210 void (*clear_block)(DCTELEM *block/*align 16*/);
211 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
212 int (*pix_sum)(uint8_t * pix, int line_size);
213 int (*pix_norm1)(uint8_t * pix, int line_size);
214 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
216 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
218 me_cmp_func hadamard8_diff[5];
219 me_cmp_func dct_sad[5];
220 me_cmp_func quant_psnr[5];
228 me_cmp_func dct_max[5];
229 me_cmp_func dct264_sad[5];
231 me_cmp_func me_pre_cmp[5];
232 me_cmp_func me_cmp[5];
233 me_cmp_func me_sub_cmp[5];
234 me_cmp_func mb_cmp[5];
235 me_cmp_func ildct_cmp[5]; //only width 16 used
236 me_cmp_func frame_skip_cmp[5]; //only width 8 used
238 int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
242 * Halfpel motion compensation with rounding (a+b+1)>>1.
243 * this is an array[4][4] of motion compensation functions for 4
244 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
245 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
246 * @param block destination where the result is stored
247 * @param pixels source
248 * @param line_size number of bytes in a horizontal line of block
251 op_pixels_func put_pixels_tab[4][4];
254 * Halfpel motion compensation with rounding (a+b+1)>>1.
255 * This is an array[4][4] of motion compensation functions for 4
256 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
257 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
258 * @param block destination into which the result is averaged (a+b+1)>>1
259 * @param pixels source
260 * @param line_size number of bytes in a horizontal line of block
263 op_pixels_func avg_pixels_tab[4][4];
266 * Halfpel motion compensation with no rounding (a+b)>>1.
267 * this is an array[2][4] of motion compensation functions for 2
268 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
269 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
270 * @param block destination where the result is stored
271 * @param pixels source
272 * @param line_size number of bytes in a horizontal line of block
275 op_pixels_func put_no_rnd_pixels_tab[4][4];
278 * Halfpel motion compensation with no rounding (a+b)>>1.
279 * this is an array[2][4] of motion compensation functions for 2
280 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
281 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
282 * @param block destination into which the result is averaged (a+b)>>1
283 * @param pixels source
284 * @param line_size number of bytes in a horizontal line of block
287 op_pixels_func avg_no_rnd_pixels_tab[4][4];
289 void (*put_no_rnd_pixels_l2[2])(uint8_t *block/*align width (8 or 16)*/, const uint8_t *a/*align 1*/, const uint8_t *b/*align 1*/, int line_size, int h);
292 * Thirdpel motion compensation with rounding (a+b+1)>>1.
293 * this is an array[12] of motion compensation functions for the 9 thirdpe
295 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
296 * @param block destination where the result is stored
297 * @param pixels source
298 * @param line_size number of bytes in a horizontal line of block
301 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
302 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
304 qpel_mc_func put_qpel_pixels_tab[2][16];
305 qpel_mc_func avg_qpel_pixels_tab[2][16];
306 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
307 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
308 qpel_mc_func put_mspel_pixels_tab[8];
313 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
314 /* This is really one func used in VC-1 decoding */
315 h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
316 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
318 qpel_mc_func put_h264_qpel_pixels_tab[4][16];
319 qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
321 qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
322 qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
324 h264_weight_func weight_h264_pixels_tab[10];
325 h264_biweight_func biweight_h264_pixels_tab[10];
328 qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
329 qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
330 void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
331 void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
332 void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
333 void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
334 void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
336 me_cmp_func pix_abs[2][4];
338 /* huffyuv specific */
339 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
340 void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w);
341 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
343 * subtract huffyuv's variant of median prediction
344 * note, this might read from src1[-1], src2[-1]
346 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
347 /* this might write to dst[w] */
348 void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp);
349 void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
351 void (*h264_v_loop_filter_luma)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0);
352 void (*h264_h_loop_filter_luma)(uint8_t *pix/*align 4 */, int stride, int alpha, int beta, int8_t *tc0);
353 /* v/h_loop_filter_luma_intra: align 16 */
354 void (*h264_v_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
355 void (*h264_h_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta);
356 void (*h264_v_loop_filter_chroma)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0);
357 void (*h264_h_loop_filter_chroma)(uint8_t *pix/*align 4*/, int stride, int alpha, int beta, int8_t *tc0);
358 void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
359 void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta);
360 // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
361 void (*h264_loop_filter_strength)(int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
362 int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field);
364 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
365 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
367 void (*h261_loop_filter)(uint8_t *src, int stride);
369 void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale);
370 void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale);
372 void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values);
373 void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values);
375 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
376 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
377 void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
378 /* no alignment needed */
379 void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
380 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
381 void (*vector_fmul)(float *dst, const float *src, int len);
382 void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
383 /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
384 void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
385 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
386 void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len);
387 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
388 void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len);
390 /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
391 * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
392 void (*float_to_int16)(int16_t *dst, const float *src, long len);
393 void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels);
396 void (*fdct)(DCTELEM *block/* align 16*/);
397 void (*fdct248)(DCTELEM *block/* align 16*/);
400 void (*idct)(DCTELEM *block/* align 16*/);
403 * block -> idct -> clip to unsigned 8 bit -> dest.
404 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
405 * @param line_size size in bytes of a horizontal line of dest
407 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
410 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
411 * @param line_size size in bytes of a horizontal line of dest
413 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
416 * idct input permutation.
417 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
419 * this permutation must be performed before the idct_put/add, note, normally this can be merged
420 * with the zigzag/alternate scan<br>
421 * an example to avoid confusion:
422 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
423 * - (x -> referece dct -> reference idct -> x)
424 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
425 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
427 uint8_t idct_permutation[64];
428 int idct_permutation_type;
429 #define FF_NO_IDCT_PERM 1
430 #define FF_LIBMPEG2_IDCT_PERM 2
431 #define FF_SIMPLE_IDCT_PERM 3
432 #define FF_TRANSPOSE_IDCT_PERM 4
433 #define FF_PARTTRANS_IDCT_PERM 5
434 #define FF_SSE2_IDCT_PERM 6
436 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
437 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
438 #define BASIS_SHIFT 16
439 #define RECON_SHIFT 6
441 void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w);
442 #define EDGE_WIDTH 16
445 /* NOTE!!! if you implement any of h264_idct8_add, h264_idct8_add4 then you must implement all of them
446 NOTE!!! if you implement any of h264_idct_add, h264_idct_add16, h264_idct_add16intra, h264_idct_add8 then you must implement all of them
447 The reason for above, is that no 2 out of one list may use a different permutation.
449 void (*h264_idct_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
450 void (*h264_idct8_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
451 void (*h264_idct_dc_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride);
452 void (*h264_idct8_dc_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride);
453 void (*h264_dct)(DCTELEM block[4][4]);
454 void (*h264_idct_add16)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
455 void (*h264_idct8_add4)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
456 void (*h264_idct_add8)(uint8_t **dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
457 void (*h264_idct_add16intra)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
460 void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
461 void (*horizontal_compose97i)(IDWTELEM *b, int width);
462 void (*inner_add_yblock)(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
464 void (*prefetch)(void *mem, int stride, int h);
466 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
469 void (*vc1_inv_trans_8x8)(DCTELEM *b);
470 void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block);
471 void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block);
472 void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block);
473 void (*vc1_v_overlap)(uint8_t* src, int stride);
474 void (*vc1_h_overlap)(uint8_t* src, int stride);
475 /* put 8x8 block with bicubic interpolation and quarterpel precision
476 * last argument is actually round value instead of height
478 op_pixels_func put_vc1_mspel_pixels_tab[16];
480 /* intrax8 functions */
481 void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize);
482 void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize,
483 int * range, int * sum, int edges);
487 * Add contents of the second vector to the first one.
488 * @param len length of vectors, should be multiple of 16
490 void (*add_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
492 * Add contents of the second vector to the first one.
493 * @param len length of vectors, should be multiple of 16
495 void (*sub_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len);
497 * Calculate scalar product of two vectors.
498 * @param len length of vectors, should be multiple of 16
499 * @param shift number of bits to discard from product
501 int32_t (*scalarproduct_int16)(int16_t *v1, int16_t *v2/*align 16*/, int len, int shift);
504 qpel_mc_func put_rv30_tpel_pixels_tab[4][16];
505 qpel_mc_func avg_rv30_tpel_pixels_tab[4][16];
508 qpel_mc_func put_rv40_qpel_pixels_tab[4][16];
509 qpel_mc_func avg_rv40_qpel_pixels_tab[4][16];
510 h264_chroma_mc_func put_rv40_chroma_pixels_tab[3];
511 h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3];
514 void dsputil_static_init(void);
515 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
517 int ff_check_alignment(void);
520 * permute block according to permuatation.
521 * @param last last non zero element in scantable order
523 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
525 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
527 #define BYTE_VEC32(c) ((c)*0x01010101UL)
529 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
531 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
534 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
536 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
539 static inline int get_penalty_factor(int lambda, int lambda2, int type){
543 return lambda>>FF_LAMBDA_SHIFT;
545 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
547 return (4*lambda)>>(FF_LAMBDA_SHIFT);
549 return (2*lambda)>>(FF_LAMBDA_SHIFT);
552 return (2*lambda)>>FF_LAMBDA_SHIFT;
557 return lambda2>>FF_LAMBDA_SHIFT;
565 * this must be called between any dsp function and float/double code.
566 * for example sin(); dsp->idct_put(); emms_c(); cos()
570 /* should be defined by architectures supporting
571 one or more MultiMedia extension */
572 int mm_support(void);
574 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
575 void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
576 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
577 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
578 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
579 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
580 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
581 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
582 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
584 #define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
586 #if defined(HAVE_MMX)
592 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
593 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
594 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
596 static inline void emms(void)
598 __asm__ volatile ("emms;":::"memory");
604 if (mm_flags & FF_MM_MMX)\
608 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
610 #elif defined(ARCH_ARM)
615 # define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
616 # define STRIDE_ALIGN 16
619 #elif defined(ARCH_PPC)
623 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
624 #define STRIDE_ALIGN 16
626 #elif defined(HAVE_MMI)
628 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
629 #define STRIDE_ALIGN 16
634 #define mm_support() 0
638 #ifndef DECLARE_ALIGNED_8
639 # define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
643 # define STRIDE_ALIGN 8
647 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
648 int orig_linesize[3], int coded_linesize,
649 AVCodecContext *avctx);
651 /* FFT computation */
653 /* NOTE: soon integer code will be added, so you must use the
655 typedef float FFTSample;
659 typedef struct FFTComplex {
663 typedef struct FFTContext {
668 FFTComplex *exptab1; /* only used by SSE code */
670 void (*fft_permute)(struct FFTContext *s, FFTComplex *z);
671 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
672 void (*imdct_calc)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
673 void (*imdct_half)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
676 int ff_fft_init(FFTContext *s, int nbits, int inverse);
677 void ff_fft_permute_c(FFTContext *s, FFTComplex *z);
678 void ff_fft_permute_sse(FFTContext *s, FFTComplex *z);
679 void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
680 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
681 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
682 void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
683 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
685 static inline void ff_fft_permute(FFTContext *s, FFTComplex *z)
687 s->fft_permute(s, z);
689 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
693 void ff_fft_end(FFTContext *s);
695 /* MDCT computation */
697 typedef struct MDCTContext {
698 int n; /* size of MDCT (i.e. number of input data * 2) */
699 int nbits; /* n = 2^nbits */
700 /* pre/post rotation tables */
706 static inline void ff_imdct_calc(MDCTContext *s, FFTSample *output, const FFTSample *input)
708 s->fft.imdct_calc(s, output, input);
710 static inline void ff_imdct_half(MDCTContext *s, FFTSample *output, const FFTSample *input)
712 s->fft.imdct_half(s, output, input);
716 * Generate a Kaiser-Bessel Derived Window.
717 * @param window pointer to half window
718 * @param alpha determines window shape
719 * @param n size of half window
721 void ff_kbd_window_init(float *window, float alpha, int n);
724 * Generate a sine window.
725 * @param window pointer to half window
726 * @param n size of half window
728 void ff_sine_window_init(float *window, int n);
729 extern float ff_sine_128 [ 128];
730 extern float ff_sine_256 [ 256];
731 extern float ff_sine_512 [ 512];
732 extern float ff_sine_1024[1024];
733 extern float ff_sine_2048[2048];
734 extern float *ff_sine_windows[5];
736 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
737 void ff_imdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
738 void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
739 void ff_imdct_calc_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
740 void ff_imdct_half_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
741 void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
742 void ff_imdct_half_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
743 void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
744 void ff_imdct_half_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
745 void ff_mdct_calc(MDCTContext *s, FFTSample *out, const FFTSample *input);
746 void ff_mdct_end(MDCTContext *s);
748 #define WRAPPER8_16(name8, name16)\
749 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
750 return name8(s, dst , src , stride, h)\
751 +name8(s, dst+8 , src+8 , stride, h);\
754 #define WRAPPER8_16_SQ(name8, name16)\
755 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
757 score +=name8(s, dst , src , stride, 8);\
758 score +=name8(s, dst+8 , src+8 , stride, 8);\
762 score +=name8(s, dst , src , stride, 8);\
763 score +=name8(s, dst+8 , src+8 , stride, 8);\
769 static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
774 AV_WN16(dst , AV_RN16(src ));
780 static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
785 AV_WN32(dst , AV_RN32(src ));
791 static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
796 AV_WN32(dst , AV_RN32(src ));
797 AV_WN32(dst+4 , AV_RN32(src+4 ));
803 static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
808 AV_WN32(dst , AV_RN32(src ));
809 AV_WN32(dst+4 , AV_RN32(src+4 ));
816 static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
821 AV_WN32(dst , AV_RN32(src ));
822 AV_WN32(dst+4 , AV_RN32(src+4 ));
823 AV_WN32(dst+8 , AV_RN32(src+8 ));
824 AV_WN32(dst+12, AV_RN32(src+12));
830 static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
835 AV_WN32(dst , AV_RN32(src ));
836 AV_WN32(dst+4 , AV_RN32(src+4 ));
837 AV_WN32(dst+8 , AV_RN32(src+8 ));
838 AV_WN32(dst+12, AV_RN32(src+12));
845 #endif /* AVCODEC_DSPUTIL_H */